Detection of damaged mooring line based on deep neural networks. (1st August 2020)

• Record Type: Journal Article
• Authors: Chung, Minwoong; Kim, Seungjun; Lee, Kanghyeok; Shin, Do Hyoung
• Is Part Of: Ocean engineering. Volume 209(2020)
• Publication Date: 2020-08-01
• Subjects: Damage detection -- Mooring line -- Tension leg platform -- Tendon -- Deep neural networks; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2020.107522
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: Since severe damage to the floating offshore structures due to the deterioration of their structural stability may lead to major disasters, it is necessary to detect mooring line damage at an early stage. However, most of the existing damage detection approaches of mooring line have difficulties to provide constant monitoring or to detect local damages to line. This study aims to develop a detection approach of a damaged mooring line in tension leg platform (TLP) based on deep neural networks (DNN). Simulation data with Charm3D was used for training and testing the DNN in the study because it is impractical to obtain actual data by intentionally damaging mooring lines that are in operation. The accuracies of the DNN model were significantly high (94.6%–99.3%) with noise level differing from 0% to 20%. The quite low false negative (FN) errors of 0.7%–5.4% for noise levels of 1–20% shows the potential of DNN-based structural health monitoring system to identify a damaged mooring line in TLP. The results of the study indicate that DNN-based damage detection approach with floater responses is applicable for even a local damage, and thus can prevent further damage or accident by early-stage detection. Highlights: This study aims to detect a damaged mooring line in tension leg platform with DNN. Simulation data of floater from CHARM3D was used for training and testing the DNN. The accuracies of the DNN models were significantly high (94.6%–99.3%). The false negative errors of models were satisfactorily low (0.7%–5.4%). DNN-based damage detection with floater responses is applicable for a local damage.